Sparse decomposition of gene expression data to infer transcriptional modules guided by motif information

An important topic in computational biology is to identify transcriptional modules through sequence analysis and gene expression profiling. A transcriptional module is formed by a group of genes under control of one or several transcription factors (TFs) that bind to cis-regulatory elements in the promoter regions of those genes. In this paper, we develop an integrative approach, namely motif-guided sparse decomposition (mSD), to uncover transcriptional modules by combining motif information and gene expression data. The method exploits the interplay of co-expression and co-regulation to find regulated gene patterns guided by TF binding information. Specifically, a motif-guided clustering method is first developed to estimate transcription factor binding activities (TFBAs); sparse component analysis is then followed to further identify TFs' target genes. The experimental results show that the mSD approach can successfully help uncover condition-specific transcriptional modules that may have important implications in endocrine therapy of breast cancer.